1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * This code is used on x86_64 to create page table identity mappings on 4 * demand by building up a new set of page tables (or appending to the 5 * existing ones), and then switching over to them when ready. 6 * 7 * Copyright (C) 2015-2016 Yinghai Lu 8 * Copyright (C) 2016 Kees Cook 9 */ 10 11 /* 12 * Since we're dealing with identity mappings, physical and virtual 13 * addresses are the same, so override these defines which are ultimately 14 * used by the headers in misc.h. 15 */ 16 #define __pa(x) ((unsigned long)(x)) 17 #define __va(x) ((void *)((unsigned long)(x))) 18 19 /* No PAGE_TABLE_ISOLATION support needed either: */ 20 #undef CONFIG_PAGE_TABLE_ISOLATION 21 22 #include "error.h" 23 #include "misc.h" 24 25 /* These actually do the work of building the kernel identity maps. */ 26 #include <linux/pgtable.h> 27 #include <asm/cmpxchg.h> 28 #include <asm/trap_pf.h> 29 #include <asm/trapnr.h> 30 #include <asm/init.h> 31 /* Use the static base for this part of the boot process */ 32 #undef __PAGE_OFFSET 33 #define __PAGE_OFFSET __PAGE_OFFSET_BASE 34 #include "../../mm/ident_map.c" 35 36 #define _SETUP 37 #include <asm/setup.h> /* For COMMAND_LINE_SIZE */ 38 #undef _SETUP 39 40 extern unsigned long get_cmd_line_ptr(void); 41 42 /* Used by PAGE_KERN* macros: */ 43 pteval_t __default_kernel_pte_mask __read_mostly = ~0; 44 45 /* Used to track our page table allocation area. */ 46 struct alloc_pgt_data { 47 unsigned char *pgt_buf; 48 unsigned long pgt_buf_size; 49 unsigned long pgt_buf_offset; 50 }; 51 52 /* 53 * Allocates space for a page table entry, using struct alloc_pgt_data 54 * above. Besides the local callers, this is used as the allocation 55 * callback in mapping_info below. 56 */ 57 static void *alloc_pgt_page(void *context) 58 { 59 struct alloc_pgt_data *pages = (struct alloc_pgt_data *)context; 60 unsigned char *entry; 61 62 /* Validate there is space available for a new page. */ 63 if (pages->pgt_buf_offset >= pages->pgt_buf_size) { 64 debug_putstr("out of pgt_buf in " __FILE__ "!?\n"); 65 debug_putaddr(pages->pgt_buf_offset); 66 debug_putaddr(pages->pgt_buf_size); 67 return NULL; 68 } 69 70 entry = pages->pgt_buf + pages->pgt_buf_offset; 71 pages->pgt_buf_offset += PAGE_SIZE; 72 73 return entry; 74 } 75 76 /* Used to track our allocated page tables. */ 77 static struct alloc_pgt_data pgt_data; 78 79 /* The top level page table entry pointer. */ 80 static unsigned long top_level_pgt; 81 82 phys_addr_t physical_mask = (1ULL << __PHYSICAL_MASK_SHIFT) - 1; 83 84 /* 85 * Mapping information structure passed to kernel_ident_mapping_init(). 86 * Due to relocation, pointers must be assigned at run time not build time. 87 */ 88 static struct x86_mapping_info mapping_info; 89 90 /* 91 * Adds the specified range to the identity mappings. 92 */ 93 void kernel_add_identity_map(unsigned long start, unsigned long end) 94 { 95 int ret; 96 97 /* Align boundary to 2M. */ 98 start = round_down(start, PMD_SIZE); 99 end = round_up(end, PMD_SIZE); 100 if (start >= end) 101 return; 102 103 /* Build the mapping. */ 104 ret = kernel_ident_mapping_init(&mapping_info, (pgd_t *)top_level_pgt, start, end); 105 if (ret) 106 error("Error: kernel_ident_mapping_init() failed\n"); 107 } 108 109 /* Locates and clears a region for a new top level page table. */ 110 void initialize_identity_maps(void *rmode) 111 { 112 unsigned long cmdline; 113 struct setup_data *sd; 114 115 /* Exclude the encryption mask from __PHYSICAL_MASK */ 116 physical_mask &= ~sme_me_mask; 117 118 /* Init mapping_info with run-time function/buffer pointers. */ 119 mapping_info.alloc_pgt_page = alloc_pgt_page; 120 mapping_info.context = &pgt_data; 121 mapping_info.page_flag = __PAGE_KERNEL_LARGE_EXEC | sme_me_mask; 122 mapping_info.kernpg_flag = _KERNPG_TABLE; 123 124 /* 125 * It should be impossible for this not to already be true, 126 * but since calling this a second time would rewind the other 127 * counters, let's just make sure this is reset too. 128 */ 129 pgt_data.pgt_buf_offset = 0; 130 131 /* 132 * If we came here via startup_32(), cr3 will be _pgtable already 133 * and we must append to the existing area instead of entirely 134 * overwriting it. 135 * 136 * With 5-level paging, we use '_pgtable' to allocate the p4d page table, 137 * the top-level page table is allocated separately. 138 * 139 * p4d_offset(top_level_pgt, 0) would cover both the 4- and 5-level 140 * cases. On 4-level paging it's equal to 'top_level_pgt'. 141 */ 142 top_level_pgt = read_cr3_pa(); 143 if (p4d_offset((pgd_t *)top_level_pgt, 0) == (p4d_t *)_pgtable) { 144 pgt_data.pgt_buf = _pgtable + BOOT_INIT_PGT_SIZE; 145 pgt_data.pgt_buf_size = BOOT_PGT_SIZE - BOOT_INIT_PGT_SIZE; 146 memset(pgt_data.pgt_buf, 0, pgt_data.pgt_buf_size); 147 } else { 148 pgt_data.pgt_buf = _pgtable; 149 pgt_data.pgt_buf_size = BOOT_PGT_SIZE; 150 memset(pgt_data.pgt_buf, 0, pgt_data.pgt_buf_size); 151 top_level_pgt = (unsigned long)alloc_pgt_page(&pgt_data); 152 } 153 154 /* 155 * New page-table is set up - map the kernel image, boot_params and the 156 * command line. The uncompressed kernel requires boot_params and the 157 * command line to be mapped in the identity mapping. Map them 158 * explicitly here in case the compressed kernel does not touch them, 159 * or does not touch all the pages covering them. 160 */ 161 kernel_add_identity_map((unsigned long)_head, (unsigned long)_end); 162 boot_params = rmode; 163 kernel_add_identity_map((unsigned long)boot_params, (unsigned long)(boot_params + 1)); 164 cmdline = get_cmd_line_ptr(); 165 kernel_add_identity_map(cmdline, cmdline + COMMAND_LINE_SIZE); 166 167 /* 168 * Also map the setup_data entries passed via boot_params in case they 169 * need to be accessed by uncompressed kernel via the identity mapping. 170 */ 171 sd = (struct setup_data *)boot_params->hdr.setup_data; 172 while (sd) { 173 unsigned long sd_addr = (unsigned long)sd; 174 175 kernel_add_identity_map(sd_addr, sd_addr + sizeof(*sd) + sd->len); 176 sd = (struct setup_data *)sd->next; 177 } 178 179 sev_prep_identity_maps(top_level_pgt); 180 181 /* Load the new page-table. */ 182 write_cr3(top_level_pgt); 183 } 184 185 static pte_t *split_large_pmd(struct x86_mapping_info *info, 186 pmd_t *pmdp, unsigned long __address) 187 { 188 unsigned long page_flags; 189 unsigned long address; 190 pte_t *pte; 191 pmd_t pmd; 192 int i; 193 194 pte = (pte_t *)info->alloc_pgt_page(info->context); 195 if (!pte) 196 return NULL; 197 198 address = __address & PMD_MASK; 199 /* No large page - clear PSE flag */ 200 page_flags = info->page_flag & ~_PAGE_PSE; 201 202 /* Populate the PTEs */ 203 for (i = 0; i < PTRS_PER_PMD; i++) { 204 set_pte(&pte[i], __pte(address | page_flags)); 205 address += PAGE_SIZE; 206 } 207 208 /* 209 * Ideally we need to clear the large PMD first and do a TLB 210 * flush before we write the new PMD. But the 2M range of the 211 * PMD might contain the code we execute and/or the stack 212 * we are on, so we can't do that. But that should be safe here 213 * because we are going from large to small mappings and we are 214 * also the only user of the page-table, so there is no chance 215 * of a TLB multihit. 216 */ 217 pmd = __pmd((unsigned long)pte | info->kernpg_flag); 218 set_pmd(pmdp, pmd); 219 /* Flush TLB to establish the new PMD */ 220 write_cr3(top_level_pgt); 221 222 return pte + pte_index(__address); 223 } 224 225 static void clflush_page(unsigned long address) 226 { 227 unsigned int flush_size; 228 char *cl, *start, *end; 229 230 /* 231 * Hardcode cl-size to 64 - CPUID can't be used here because that might 232 * cause another #VC exception and the GHCB is not ready to use yet. 233 */ 234 flush_size = 64; 235 start = (char *)(address & PAGE_MASK); 236 end = start + PAGE_SIZE; 237 238 /* 239 * First make sure there are no pending writes on the cache-lines to 240 * flush. 241 */ 242 asm volatile("mfence" : : : "memory"); 243 244 for (cl = start; cl != end; cl += flush_size) 245 clflush(cl); 246 } 247 248 static int set_clr_page_flags(struct x86_mapping_info *info, 249 unsigned long address, 250 pteval_t set, pteval_t clr) 251 { 252 pgd_t *pgdp = (pgd_t *)top_level_pgt; 253 p4d_t *p4dp; 254 pud_t *pudp; 255 pmd_t *pmdp; 256 pte_t *ptep, pte; 257 258 /* 259 * First make sure there is a PMD mapping for 'address'. 260 * It should already exist, but keep things generic. 261 * 262 * To map the page just read from it and fault it in if there is no 263 * mapping yet. kernel_add_identity_map() can't be called here because 264 * that would unconditionally map the address on PMD level, destroying 265 * any PTE-level mappings that might already exist. Use assembly here 266 * so the access won't be optimized away. 267 */ 268 asm volatile("mov %[address], %%r9" 269 :: [address] "g" (*(unsigned long *)address) 270 : "r9", "memory"); 271 272 /* 273 * The page is mapped at least with PMD size - so skip checks and walk 274 * directly to the PMD. 275 */ 276 p4dp = p4d_offset(pgdp, address); 277 pudp = pud_offset(p4dp, address); 278 pmdp = pmd_offset(pudp, address); 279 280 if (pmd_large(*pmdp)) 281 ptep = split_large_pmd(info, pmdp, address); 282 else 283 ptep = pte_offset_kernel(pmdp, address); 284 285 if (!ptep) 286 return -ENOMEM; 287 288 /* 289 * Changing encryption attributes of a page requires to flush it from 290 * the caches. 291 */ 292 if ((set | clr) & _PAGE_ENC) { 293 clflush_page(address); 294 295 /* 296 * If the encryption attribute is being cleared, change the page state 297 * to shared in the RMP table. 298 */ 299 if (clr) 300 snp_set_page_shared(__pa(address & PAGE_MASK)); 301 } 302 303 /* Update PTE */ 304 pte = *ptep; 305 pte = pte_set_flags(pte, set); 306 pte = pte_clear_flags(pte, clr); 307 set_pte(ptep, pte); 308 309 /* 310 * If the encryption attribute is being set, then change the page state to 311 * private in the RMP entry. The page state change must be done after the PTE 312 * is updated. 313 */ 314 if (set & _PAGE_ENC) 315 snp_set_page_private(__pa(address & PAGE_MASK)); 316 317 /* Flush TLB after changing encryption attribute */ 318 write_cr3(top_level_pgt); 319 320 return 0; 321 } 322 323 int set_page_decrypted(unsigned long address) 324 { 325 return set_clr_page_flags(&mapping_info, address, 0, _PAGE_ENC); 326 } 327 328 int set_page_encrypted(unsigned long address) 329 { 330 return set_clr_page_flags(&mapping_info, address, _PAGE_ENC, 0); 331 } 332 333 int set_page_non_present(unsigned long address) 334 { 335 return set_clr_page_flags(&mapping_info, address, 0, _PAGE_PRESENT); 336 } 337 338 static void do_pf_error(const char *msg, unsigned long error_code, 339 unsigned long address, unsigned long ip) 340 { 341 error_putstr(msg); 342 343 error_putstr("\nError Code: "); 344 error_puthex(error_code); 345 error_putstr("\nCR2: 0x"); 346 error_puthex(address); 347 error_putstr("\nRIP relative to _head: 0x"); 348 error_puthex(ip - (unsigned long)_head); 349 error_putstr("\n"); 350 351 error("Stopping.\n"); 352 } 353 354 void do_boot_page_fault(struct pt_regs *regs, unsigned long error_code) 355 { 356 unsigned long address = native_read_cr2(); 357 unsigned long end; 358 bool ghcb_fault; 359 360 ghcb_fault = sev_es_check_ghcb_fault(address); 361 362 address &= PMD_MASK; 363 end = address + PMD_SIZE; 364 365 /* 366 * Check for unexpected error codes. Unexpected are: 367 * - Faults on present pages 368 * - User faults 369 * - Reserved bits set 370 */ 371 if (error_code & (X86_PF_PROT | X86_PF_USER | X86_PF_RSVD)) 372 do_pf_error("Unexpected page-fault:", error_code, address, regs->ip); 373 else if (ghcb_fault) 374 do_pf_error("Page-fault on GHCB page:", error_code, address, regs->ip); 375 376 /* 377 * Error code is sane - now identity map the 2M region around 378 * the faulting address. 379 */ 380 kernel_add_identity_map(address, end); 381 } 382